On 22/02/2023 9:53 am, Jan Beulich wrote:
> On 22.02.2023 10:42, Jan Beulich wrote:
>> On 21.02.2023 19:05, Andrew Cooper wrote:
>>> On 21/02/2023 4:55 pm, Anthony PERARD wrote:
>>>> Building randconfig on debian unstable seems to be an issue.
>>> You're talking about
>>> https://gitlab.com/xen-project/people/anthonyper/xen/-/jobs/3769926509 ?
>>>
>>> + gcc --version
>>> gcc (Debian 12.2.0-14) 12.2.0
>>>
>>> arch/x86/extable.c: In function 'search_pre_exception_table':
>>> arch/x86/extable.c:200:27: error: array subscript -1 is outside array
>>> bounds of 'struct exception_table_entry[1152921504606846975]'
>>> [-Werror=array-bounds]
>>> 200 | unsigned long fixup = search_one_extable(
>>> | ^~~~~~~~~~~~~~~~~~~
>>> 201 | __start___pre_ex_table, __stop___pre_ex_table-1, addr);
>>> | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
>>> In file included from arch/x86/extable.c:8:
>>> ./arch/x86/include/asm/uaccess.h:414:37: note: at offset -8 into object
>>> '__stop___pre_ex_table' of size [0, 9223372036854775807]
>>> 414 | extern struct exception_table_entry __stop___pre_ex_table[];
>>> | ^~~~~~~~~~~~~~~~~~~~~
>>> cc1: all warnings being treated as errors
>>> make[3]: *** [Rules.mk:246: arch/x86/extable.o] Error 1
>>>
>>> Jan: do we need some more gcc-wrap sprinkled around?
>> I think we can get away without, but first of all I'm curious why gcc12
>> spots an issue there, but not in my (and perhaps also your) local builds.
>> Being a randconfig failure, is this tied to a specific setting in .config?
>> Off the top of my head I can't think of any which might affect this code.
> Looks to be the combination of DEBUG=n and COVERAGE=y.
>
>> To work around this, instead of adding gcc-wrap-ery, I'd favor simply
>> changing search_one_extable()'s 2nd parameter's meaning from "last" to
>> "end" (i.e. the range passed in from [,] to [,) in mathematical spelling).
> I'll make a patch along these lines then, I guess. Should I credit Anthony,
> you, or both of you as the reporter?
Yeah, I did wonder about changing the final parameter after sending this
email. It stands a reasonable change of simplifying
search_one_extable() anyway.
Probably Anthony. All I did was look through some logs.
~Andrew